Neuroimaging Markers for Studying Gulf-War Illness: Single-Subject Level Analytical Method Based on Machine Learning

• Published in: Brain sciences Vol. 10; no. 11; p. 884
• Main Authors: Guan, Yi, Cheng, Chia-Hsin, Chen, Weifan, Zhang, Yingqi, Koo, Sophia, Krengel, Maxine, Janulewicz, Patricia, Toomey, Rosemary, Yang, Ehwa, Bhadelia, Rafeeque, Steele, Lea, Kim, Jae-Hun, Sullivan, Kimberly, Koo, Bang-Bon
• Format: Journal Article
• Language: English
• Published: Basel MDPI AG 01.11.2020; MDPI
• Subjects: Biological markers; Biomarkers; Classification; Consortia; Diagnosis; diffusion; Fatigue; Gulf War; Gulf War illness; Gulf War syndrome; Health aspects; Illnesses; Kansas case criteria; Learning algorithms; Machine learning; Magnetic resonance imaging; Medical imaging; Military personnel; MRI; Neuroimaging; objective biomarker; Pain; Persian Gulf syndrome; Persian Gulf War; Questionnaires; Registration; Sleep; Substantia alba; Technology application; Traumatic brain injury; War; United States; United States > US; Kansas
• ISSN: 2076-3425; 2076-3425
• DOI: 10.3390/brainsci10110884

Gulf War illness (GWI) refers to the multitude of chronic health symptoms, spanning from fatigue, musculoskeletal pain, and neurological complaints to respiratory, gastrointestinal, and dermatologic symptoms experienced by about 250,000 GW veterans who served in the 1991 Gulf War (GW). Longitudinal studies showed that the severity of these symptoms often remain unchanged even years after the GW, and these veterans with GWI continue to have poorer general health and increased chronic medical conditions than their non-deployed counterparts. For better management and treatment of this condition, there is an urgent need for developing objective biomarkers that can help with simple and accurate diagnosis of GWI. In this study, we applied multiple neuroimaging techniques, including T1-weighted magnetic resonance imaging (T1W-MRI), diffusion tensor imaging (DTI), and novel neurite density imaging (NDI) to perform both a group-level statistical comparison and a single-subject level machine learning (ML) analysis to identify diagnostic imaging features of GWI. Our results supported NDI as the most sensitive in defining GWI characteristics. In particular, our classifier trained with white matter NDI features achieved an accuracy of 90% and F-score of 0.941 for classifying GWI cases from controls after the cross-validation. These results are consistent with our previous study which suggests that NDI measures are sensitive to the microstructural and macrostructural changes in the brain of veterans with GWI, which can be valuable for designing better diagnosis method and treatment efficacy studies.